Szu-Wei Fu comments

Results 17 comments of


                                            Szu-Wei Fu

~loss: Nan

HI, I never met this nan problem before. I'm not sure whether it is due to a different python version (mine is Python 2.7) or PESQ code. Maybe you can...

Why consider the filter number 30 and 55 as the convolution window length?

We gradually increased the length and number of filters and found it works well on the validation set.

Batch Size = 1

hi, Because different utterances have different lengths, to use Batch Size >1, input padding is required. We also tried to use Batch Size >1 in a Pytorch version, but the...

which cuda and cudnn version did you use?

Sorry, my server is out of work now, so I cannot get the related information. Try to update your GPU driver version?

Question on Weight Calculation

There are 30 channels in each layer (except for the input and output layer), so the number of weights should be roughly multiplied with 30.

method

Hi Qiquan, 1) After checking model.summary(), the number of parameters for G and D are 1,892,057 and 690,855, respectively. 2) Yes, it is non-causal. Sincerely, Szu-Wei

Hi, I guess you are running the dataset used in table2. Please use https://github.com/JasonSWFu/MetricGAN/blob/master/pesq_cd.m for PESQ evaluation. Please also note that, as mentioned in the paper, the input features and...

评价指标

Hi, as mentioned in the paper, the input features and activation functions used in table 2 are different from those provided here. To easily get improved results, you can try...